JPH03260803A - Generating method for fuzzy control rule - Google Patents

Abstract

PURPOSE:To automatically generate a fuzzy control rule by making use of a hierarchical network constitution data processor. CONSTITUTION:In a processing process 1, a membership function is generated for acquisition of a fuzzy control rule. In a processing process 2, a data processing means 10 containing a hierarchical network part 11 is prepared for allocation of an arithmetic function of the membership function generated in a processing layer of the precedent stage, an arithmetic function of the fuzzy control rule generated in an intermediate processing layer, and an arithmetic function of the fuzzy control rule generated in a processing layer of the next stage respectively. Then in a processing 3, the learning of an internal state value is executed, and in a processing process 4, the internal state value is learnt, and the fuzzy control rule is extracted with analysis of the learnt internal state value in a processing process 4. Thus the fuzzy control rule is automatically and mechanically obtained.

Description

[Detailed Description of the Invention] [Summary] Regarding a fuzzy control rule generation method for generating fuzzy control rules, the antecedent part and consequent part members constituting the fuzzy control rule are provided for the purpose of automatically generating fuzzy control rules. A first processing step that generates a ship function and a data processing means that includes a three-layer hierarchical network section are prepared, and the calculation function of the generated antecedent membership function is assigned to the previous processing layer, and the intermediate The antecedent part calculation functions of the fuzzy control rules for the number of generated rules are assigned to the processing layer, and the consequent part calculation functions of the fuzzy control rules for the number of generated consequent membership functions are assigned to the subsequent processing layer. 2, the control amount data calculated from the output value from the subsequent processing layer output according to the presentation of the control presentation data for learning and the generated consequent membership function is the control teacher data for learning. and a fourth processing step for extracting fuzzy control rules by analyzing the learned internal state values.

[Industrial application field]

The present invention relates to a fuzzy control rule generation method for generating fuzzy control rules required for constructing a fuzzy controller, and in particular, a method for automatically generating fuzzy control rules using a hierarchical network configuration data processing device. This invention relates to a method for generating fuzzy control rules that can be generated in a variety of ways.

Fuzzy control is becoming popular as a new control processing method. This fuzzy control expresses a control algorithm that includes ambiguity such as human judgment in an if-then format, and executes this control algorithm according to fuzzy reasoning to calculate the control operation amount from the detected control state quantity. The control target is controlled by calculating the In order to realize this fuzzy control, it is necessary to generate fuzzy control rules that describe the fuzzy control algorithm. In order to make fuzzy controllers practical, it is necessary to be able to automatically generate fuzzy control rules.

[Conventional technology]

For example, if x, is big and xz is ss
Generation of fuzzy control rules written in the format +all then y is big (the if part is called the antecedent part and the then part is called the consequent part) has traditionally been done by an operator who is familiar with controlling the controlled object. According to our knowledge of the target process, we first quantify the linguistic meaning of the values of control state quantities and control operation quantities as membership functions, and then describe the relationships between these membership functions. This generates a rough model of fuzzy control rules. This was then achieved by evaluating and tuning the generated coarse fuzzy control rule model through simulations and on-site tests to create fuzzy control rules that were suitable for the control target.

[Problem to be solved by the invention]

In this way, conventionally, generation of fuzzy control rules is
This was done manually by a collaborative effort between a designer who wanted to build a fuzzy controller and an operator who was familiar with controlling the controlled object. The work to generate this fuzzy control rule requires trial and error, making full use of know-how, so it usually takes several months at the shortest. If there is little control knowledge regarding the controlled object, this period will be even longer.

The conventional fuzzy control rule generation method has the problem of forcing the designer of the fuzzy controller and the operator of the control object to perform extremely burdensome work. Moreover, since the generated fuzzy control rules directly affect the control performance of the fuzzy controller, there is a strong desire to improve the quality of the fuzzy control rules.

The present invention has been made in view of the above circumstances, and by providing a new fuzzy control rule generation method that can automatically and mechanically generate fuzzy control rules, the amount of work required to generate fuzzy control rules can be reduced. The purpose is to make it possible to significantly reduce the amount of fuzzy control rules and to generate objective fuzzy control rules.

[Means to solve the problem]

FIG. 1 is a diagram showing the principle configuration of the present invention.

In the figure, Pl is the first treatment process, P2 is the second treatment process,
P3 is the third processing step, and P4 is the fourth processing step.

10 is a hierarchical network configuration data processing means, 11 is a hierarchical network unit included in the hierarchical network configuration data processing unit 10, 12 is an input unit forming the input layer of the hierarchical network unit 11, and 13 is a hierarchical network unit 1
1, a processing unit forming the first processing layer of the hierarchical network section 11; 14, a processing unit forming the second processing layer of the hierarchical network section 11; 15, a processing unit forming the third processing layer of the hierarchical network section 11; is the hierarchical network part 11
A processing unit 17 is an output unit forming the output layer of the hierarchical network unit 11.

In many cases, a group of processing units 13 in the first processing layer that perform arithmetic processing regarding the same control state quantity (control operation amount) is only a group of processing units 13 that perform processing on the same specific input value (control operation amount). An internally coupled configuration is adopted. The internal state value assigned to this internal connection can be excluded from learning. Processing unit 1 of the first processing layer
3 and the processing unit 14 of the second processing layer, and between the processing unit 14 of the second processing layer and the processing unit 15 of the third processing layer are mutually internally coupled. The internal state value assigned to this internal connection becomes the learning target. The processing unit 15 of the third processing layer and the processing unit 16 of the fourth processing layer are configured to be internally coupled with each other in a corresponding relationship. The internal state value assigned to this internal connection can be excluded from learning. The processing units 16 of the fourth processing layer that execute arithmetic processing regarding the same control operation amount (control state amount) are configured to be internally coupled with the corresponding output unit 17. The internal state value assigned to this internal connection can be excluded from learning.

The processing units 14 of the second processing layer may be further hierarchically configured in multiple stages, and the processing units 15 of the third processing layer may also be further hierarchically configured in multiple stages. the processing unit of the second processing layer, which may be
14 are hierarchically configured in multiple stages, a configuration is adopted in which the processing units 14 in adjacent layers are mutually internally coupled, and the processing units 15 in the third processing layer are configured in a hierarchical manner. In the case of a multi-stage configuration, a configuration is adopted in which the processing units 15 in adjacent layers are internally connected to each other.

Reference numeral 20 denotes a learning signal presentation device, in which a group of control data for learning (a pair of a control state quantity and a control operation quantity is used as a basic unit,
One of them becomes the control presentation data and the other one becomes the control teacher data which is a teacher signal.The control presentation data of the two is presented to the input layer of the hierarchical network unit 11, and the control teacher data will be explained next. What is presented to the internal state value learning processing device 30 is an internal state value learning processing device implementing, for example, a back propagation method, and output information 17 is output in accordance with the presentation of control presentation data by the learning signal presentation device 220. , and the control teacher data group presented from the learning signal presentation device 20, and sequentially update the internal state values to be learned based on the error values. By doing so, the internal state value for which the error value falls within the allowable range is learned.

In the first processing step P1, from the previous control experience for the controlled object, first of all, what kind of control state quantities exist for the controlled object, and what kind of control operation variables exist. Then, by generating a membership function for the list-amplified control state quantity and a membership function for the restored control operation quantity, we can evaluate the possibility of constructing fuzzy control rules. A certain antecedent membership function and a consequent membership function are generated.

Here, when the antecedent part of the fuzzy control rule is a description regarding the control state quantity, the membership function regarding the generated control state quantity becomes the antecedent membership function, and the member regarding the generated control operation quantity becomes the antecedent membership function. When the ship function becomes the consequent membership function and the antecedent part of the fuzzy control rule is a description regarding the control operation amount, the membership function regarding the generated control operation amount becomes the antecedent membership function, and
The membership function regarding the generated control state amount becomes the consequent membership function.

Subsequently, in the second processing step P2, the calculation function of the antecedent membership function generated in the first processing step P1 is assigned to the processing unit 13 of the first processing layer, and the second
The antecedent computation functions of the fuzzy control rules for the number of fuzzy control rules to be generated are assigned to the processing unit 14 of the processing layer, and the processing unit 15 of the third processing layer is assigned the first processing function. The consequent calculation functions of the fuzzy control rules corresponding to the number of consequent membership functions generated in process P1 are assigned, and the processing unit 1 of the fourth processing layer is
6, the calculation function of the consequent membership function generated in the first processing step P1 is assigned, and to output unit 17, the calculation function is assigned from the truth value of the consequent membership function related to the same attribute. A process of allocating an operator for calculating control amount data serving as a fuzzy inference value is executed.

The process of assigning the calculation function of the consequent membership function to the processing unit 16 of the fourth processing layer is such that all the consequent membership functions related to the same attribute are managed by one processing unit 16. When adopting a configuration that
On the other hand, when adopting a configuration in which consequent membership functions related to the same attribute are distributed and managed by multiple processing units 16, those processing It will be allocated to unit 16. In FIG. 1, a calculation function of a consequent membership function is assigned to a plurality of processing units 16.

In this second processing step P2, when the processing units 14 of the second processing layer are hierarchically configured in multiple stages, the antecedent part of the fuzzy control rule is added to the second processing layer configured in multiple stages. Arithmetic functions will be assigned. Also,
When the processing units 15 of the third processing layer are hierarchically configured in multiple stages, the consequent calculation function of the fuzzy control rule is assigned to the third processing layer configured in multiple stages.

In the subsequent third processing step P3, the learning signal presentation device 20
By activating the control presentation data for learning is presented to the processing unit 13 of the first processing layer via the input crab unit 12, and the control presentation data for learning is presented to the internal state value learning processing device 30. Present the data. Then, by activating this internal state value learning processing device 30, the control amount data from the output unit 17 that is output in accordance with the presentation of the control presentation data for learning will approximately match the control teacher data that is presented. Execute learning processing for the internal state value of the learning target as much as possible.

Subsequently, in the fourth processing step P4, the learned internal state value of the internal connection between the processing unit 13 and the processing unit 14, the internal state value of the internal connection between the processing unit 14 and the processing unit 15, In the case of a multi-stage configuration, fuzzy control rules are generated by analyzing internal state values of internal connections between processing units 14 and between processing units 15 in addition to this.

[Effect]

In the present invention, according to the process of the first processing step P1, for example, "the outlet pressure is high" regarding the outlet pressure and water level that are the control state quantities of the controlled object.

"Outlet pressure is low", "Outlet pressure is extremely low", "
In addition to generating a membership function that numerically expresses "high water level,""medium water level," and "low water level," it also generates a membership function that numerically expresses "water level is high,""water level is medium," and "water level is low." A member who numerically expresses things such as ``greatly increase the amount of heat supply'', ``reduce the amount of heat supply'', ``increase the throttle valve opening'', and ``reduce the throttle valve opening'' regarding degrees. Generate a ship function.

According to the subsequent second processing step P2, for example, the antecedent part of the fuzzy control rule to be generated describes the conditions regarding the @comb state quantity, and the consequent part describes the conditions regarding the control operation amount. Sometimes, the processing function of the membership function regarding the generated control state quantity is assigned to the processing unit 13 of the first processing layer, and the calculation function of the membership function regarding the generated control operation quantity is assigned to the processing unit 16 of the fourth processing layer. Assign output functions. That is,
Each processing unit 13 of the first processing layer is configured to calculate and output the membership function value (truth value) of the assigned membership function according to the human power of the control state quantity, and the fourth Each processing unit 16 of the processing layer
However, the truth value of the assigned membership function is output to the output unit 17 (as will be explained in the embodiment, a reduced value is output according to the output value from the processing unit 15, etc.). It composes.

Here, the processing unit) 13 that calculates the membership function for the same type of control state quantity such as outlet pressure can only be the same input unit) 12 unless the membership function is a function of multiple control state quantities. (a plurality of processing units) that are internally coupled and calculate membership functions for the same type of control operation amount such as supply heating amount operation;
6 performs allocation processing so as to internally connect only with the same output unit 17.

Furthermore, in accordance with the processing of the second processing step P2, an antecedent part function Xll function is provided to the processing unit 14 of the second processing layer, for example, to select and output the minimum value of the human power values that perform the logical product operation. At the same time, a consequent calculation function is allocated to the processing unit 15 of the third processing layer, which selects the maximum value of the input values forming the logical sum operation and outputs it to the processing unit 16.

Through this second processing step P2, all fuzzy control rules that have a possibility of being established are
It will be mapped onto the top.

When the fuzzy control rule mapping process for the hierarchical network unit 11 is completed, the actual control data group obtained from the controlled object for which the fuzzy control rule is generated is processed according to the subsequent third processing step P3. is used as a teaching signal for learning, for example, when the antecedent part of the fuzzy control rule to be generated describes a condition regarding the control state quantity, the control state quantity data in the control data group for learning is controlled. In addition to using it as presentation data,
The learning process of the internal state value of the hierarchical network unit 11 is executed using the other control operation amount data as control teacher data.

Through this learning process, the actual control contents of the controlled object are reflected in the internal state values of the hierarchical network unit 11. In other words, the internal state value of the internal connection between the antecedent membership function and the antecedent operation, which are defined as having a deeper relationship, and the antecedent operation and the consequent operation (consequent membership function) The internal state value of the internal connection between is learned to be the one with the larger value.

From now on, according to the fourth processing step P4, the internal state value that takes a large value is extracted, and the antecedent membership function and the antecedent performance that are related by the internal state value that has the large value are extracted. A fuzzy control rule is generated by performing analysis processing such as identifying a consequent operation and a consequent membership function.

Thus, according to the present invention, fuzzy control rules can be automatically and mechanically generated.

[Example] Hereinafter, the present invention will be explained in detail according to examples.

In FIG. 2, the system configuration of an apparatus used for implementing the present invention is shown. In FIG. 2, the same components as those explained in FIG. 1 are indicated by the same symbols. 1''-h is a plurality of input units forming the input layer of the hierarchical network unit 11, l-i is a plurality of basic units forming the intermediate layer (provided in multiple stages) of the hierarchical network unit 11, 1-
j is one or more basic units constituting the output layer of the hierarchical network unit 11, loa is a hierarchical network configuration data processing device forming the hierarchical network configuration data processing means lO, and 18 is a weight value storage unit, which is a hierarchical network configuration data processing unit. It manages the weight values assigned to the internal connections of the section 11.

21 is a learning signal storage unit that stores learning control data used for learning weight values; 22 is a learning signal presentation unit that stores learning control data from the learning signal storage unit 21; and presents the control presentation data therein to the hierarchical network unit 11, and the weight value changing unit 3 which will be described later the other pair of control teacher data.
0a and what is presented to a learning convergence determination unit 23, which will be described next, 23 is a learning convergence determination unit that presents control amount data output from the hierarchical network unit 11 and a learning signal presentation unit 22
It receives the control teacher data from the hierarchical network unit 11, determines whether the error in the data processing function of the hierarchical network unit 11 is within an allowable range, and notifies the learning signal presentation unit 22 of the determination result.

30a is a weight value changing unit corresponding to the internal state value learning processing device 30 in FIG. It learns to converge the weight values by setting the update amount of the weight values to zero according to the propagation method and updating the weight values according to the update amounts.

In the present invention, as explained with reference to FIG. 1, the system shown in FIG. 2 is used to construct an automatic generation process for fuzzy control rules. Before entering into a detailed explanation of the present invention, the device configuration of the hierarchical network section 11 adopted for realizing normal data processing will be explained in detail, and the weight values assigned to internal connections of this hierarchical network section 11 will be learned. In order to do this, the back propagation method (D, E, Rumelhart, G, E, H
inton, andRjJilliams, “Lea
rning Internal Representa
tions by ErrorPropagation
+PARALLEL DISTRIBUTED P
ROCESSING, Vol, l, pp, 318-
364+ The MITPress, 1986).

The hierarchical network unit 11 basically constitutes a hierarchical network from a type of node called a basic unit 1 and internal connections having weight values.

FIG. 3 shows the basic configuration of the basic unit 1. This basic unit 1 has a multi-input/output system, and includes a multiplication processing section 2 that multiplies a plurality of inputs by weight values of respective internal connections, and a multiplication processing section 3 that adds all the multiplication results. and,
It includes a dark value processing section 4 that performs slightly nonlinear threshold processing this summer and outputs one final output.

Assuming that the h layer is the first layer and the i layer is the second layer, the accumulation processing section 3 executes the calculation of the following equation (1), and the threshold processing section 4
Now, execute the calculation of equation (2> below).

Xpi-Σ)'bW=b (1) Formula y pi = 1 / (1+exp(x pt+θi
)) Formula (2) where, h : Unit number of h layer i : Unit number of i layer p : Input signal pattern number θ, : Darkness value W of the 1st unit of i layer, , : Between h-j layers Weight value of internal connection 1 of the i layer for the input signal of the th pattern
In the hierarchical network section 11, a large number of basic units 1 having such a configuration use the input unit 1', which distributes and outputs the input signal value as it is, as an input layer, as shown in FIG. By being connected hierarchically to form a hierarchical network, the parallel data processing function of converting an input pattern (input signal) into a corresponding output pattern (output signal) is realized.

In the back propagation method, the weight value Wik and threshold value θ of the hierarchical network are adaptively and automatically adjusted and learned using error feedback. (1)
As is clear from equation (2), it is necessary to adjust the weight value W1 and the threshold value θ8 at the same time, but this task is a complicated task that interferes with each other. Therefore, the applicant
As previously disclosed in Japanese Patent Application No. 1983-333484 (filed December 28, 1988, "Network configuration data processing device"), "1" is always output to the h layer at the input end. We also proposed that by providing a unit that allocates the threshold value θ as a weight value to the output, the threshold value θi is incorporated into the weight value W, and the threshold value θ is treated as a weight value. By doing so,
The above equations (1) and (2) are as follows:
s=! 1/ (1+exp(x*=)) (4
) will be expressed by the formula.

Next, according to the written format of equations (3) and (4),
The weight value learning processing method using the back propagation method will be explained. Here, this explanation will be given assuming that the hierarchical network unit 11 has a three-layered hierarchical network of layer h, layer 1, and layer j as shown in FIG.

(3) By analogy with equations (4), the following (5) (6
) formula is obtained. That is, (5) Formula % Formula % )) j: Unit number of layer j wjt: Weight value of internal connection between layers t-j xpj: Sum of products of inputs from each unit of layer j to the first unit of layer J 7pj'The output weight value changing unit 30a from the first unit of the j layer in equation (6) for the input signal of the P-th pattern calculates the output pattern y from the output layer when the human cover pattern for learning is presented. ,
When the receiver takes the teacher pattern dpj (the teacher signal to the jJij-th unit for the input signal of the p-th pattern), which is the signal that the output pattern Vei should take,
First, the difference value (dp, yp=) between the output pattern y□ and the teacher pattern d is calculated, and then α2. =y□(1-y□)(d□-yl) Formula (7) is calculated, and then ΔWJt(t)-εΣα1y□+ζΔWJt(L-1)
(8) Formula % Formula %: : According to the update amount of the weight value between the i layer and the j layer ΔW j 1
(t) is calculated. Here, “ζΔW, 1(t-1)J
The reason for adding the update amount with a small weight determined in the previous update cycle is to speed up learning.

Next, the weight value changing unit 30a first uses the calculated α and J to calculate βpi ” 7 pt (1-3' 9.) Σα, JW
,,(t-1) Calculate equation (9), then ΔW, ゎ(1)-εΣβpiy+ζΔWik(t-1
) According to equation (10), the update amount ΔW of the weight value between layer h and layer 1 is calculated.

Subsequently, the weight value changing unit 30a changes the weight value Jt(t) − for the next update cycle according to the calculated update amount.
WJi(t-1)+ΔWji(t)W! b(t) −W
By repeating the method of determining lb(t 1) + ΔWih(t), the output cover turn Yp from the output layer when the input cover turn for learning is presented
i and the teacher pattern dp1, which is the signal that the output pattern VeJ should take, match the weight value W, .
, W, and h.

Then, the hierarchical network unit 11 connects the g layer to the h layer to the i layer = j
When you have a hierarchical network with a four-layer structure called layers,
The weight value changing unit 30a firstly calculates T,,=y ph(171)Σβ,,W,,(t-
1) Calculate equation (11), then ΔWbg(t)=εΣTpb7□+ζΔW, (t-1
) According to equation (12), update amount ΔWh of weight values between g layer and h layer, (t)
In this way, the update amount ΔW of the weight value between the eyebrows in the previous stage is determined using the value obtained from the latter stage on the output side and the network output data.

Here, if the basic unit 1 does not include the threshold processing section 4, the above equation (7) becomes α□-(d□-ypi) (7) equation, and the above equation (9) becomes β □−Σα, Wit (t 1) (9), and the above equation (11) is as follows: γ, = Σβ, = Wib (t 1) (11)
A method of configuring the hierarchical network unit 11 using hardware components is described in Japanese Patent Application No. 1983-2 filed by the present applicant.
No. 16865 (filed August 31, 1988, ``Network configuration data processing device'')'' can be used.

That is, as shown in FIG. 5, the basic unit 1 is a multiplication type D that multiplies the output from the previous layer input via the input switch section 7 by the weight value held by the weight value holding section 8.
/A converter 2a, an analog adder 3a that adds the output value of the multiplication type D/A converter 2a and the previous accumulated value to calculate a new accumulated value, and holds the addition result of the analog adder 3a. A sample-and-hold circuit 3b that converts the data held in the sample-and-hold circuit 3b into nonlinear form when the accumulation process is completed, and an analog signal of the nonlinear function generation circuit 4a that outputs the data to the subsequent layer. This is realized by including an output holding section 5 that holds a value, an output switch section 6 that outputs the data held by the output holding section 5, and a control circuit 9 that controls each of these processing sections.

The hierarchical network unit 11 is realized in a configuration in which basic units 1 having this configuration are electrically connected by one common analog bus 70, as shown in FIG. Here, in the figure, 71 is a weight output circuit that gives a weight value to the weight holding section 8 of the basic unit I, 72 is an initial signal output circuit corresponding to the input unit l', and 73 is a synchronous control signal that is a control signal for data transfer. A signal weighting output circuit 71,
A synchronous control signal line 74 that transmits to the initial signal output circuit 72 and the control circuit 9 is a main control circuit that sends out a synchronous control signal.

In the hierarchical network unit 11 having this configuration, the main control circuit 74 sequentially selects the basic units 1 in the previous layer in chronological order, and in synchronization with this selection process, the output holding unit of the selected basic unit 1. 5 is processed so as to be outputted to the multiplication type D/A converter 2a of the basic unit 1 in the subsequent layer via the analog bus 70 in accordance with the time-division transmission format. When this input is received, the multiplication type D/A converter 2a of the basic unit 1 in the subsequent layer sequentially selects the corresponding weight values, performs the processing of the human power value and the weight value, and connects it to the analog adder 3a. The accumulation processing section 3 constituted by the sample and hold circuit 3b sequentially accumulates the multiplied values.

Subsequently, when all the accumulation processing for the basic unit 1 in the previous layer is completed, the main control circuit 74 activates the nonlinear function generating circuit 4a of the basic unit 1 in the subsequent layer to calculate the final output. , the output holding unit 5 processes to hold the final output of the conversion processing result. And the main control circuit 7
4, by repeating the same process for the next subsequent layer using this latter layer as a new previous layer, processing is performed so that an output cover turn corresponding to an input cover turn is output.

Next, automatic generation processing of fuzzy control rules according to the present invention, which will be realized using the system shown in FIG. 2, will be explained. First, an embodiment will be described in which a hierarchical network section 11 made up of hardware components as shown in FIGS. 5 and 6 is used.

In the present invention, a hierarchical network unit 11 constituted by hardware components as shown in FIGS. 5 and 6, for example.
The process begins by mapping all fuzzy control rules that may be adopted onto this hierarchical network unit 11 using the following. For this reason, a user who desires to generate a fuzzy control rule first generates a membership function for the control state quantity and control operation amount regarding the controlled object, based on the control experience with respect to the controlled object. That is, as before, as shown in FIG.
For example, a membership function is generated that describes the ambiguity of "temperature is high", "temperature is normal", and "temperature is low J" as numerical data.

Next, when the user generates a fuzzy control rule in which the antecedent part describes a condition regarding the control state amount and the consequent part describes a condition regarding the control operation amount, the user Using the basic unit 1 of the layer, a function for calculating and outputting the truth value of the membership function for the generated control state quantity (a function for calculating the truth value of the antecedent membership function) is assigned, and
The function of outputting the truth value of the membership function for the generated control operation amount (the function of outputting the truth value of the consequent membership function) is performed using the basic unit 1 of the intermediate layer located at the fourth position. Execute the allocation process.

Specifically, the assignment process of the truth value calculation function of the antecedent membership function is such that the output value y 3 1+exP (-ωχ+θ) output from the basic unit 1 is as shown in FIG. 8(a). To, ω ku 0. When θ〈0, the function shape is similar to the membership function ``low temperature'' in Figure 7, and when ω〉0.0〉0, the membership function ``high temperature'' in Figure 7 is obtained. Focusing on the fact that the function shape is similar to the membership function, as shown in FIG. 8(b), by appropriately setting the weight value ω and threshold value θ assigned to the input of this basic unit 1, Membership functions having the function shapes "low temperature" and "high temperature" shown in FIG. 7 are realized.

Furthermore, the difference value y 1+eXp (-(1), Focusing on the fact that it has a function shape similar to the membership function "Temperature is normal" in Figure 7, as shown in Figure 9(b), two basic units 1 and these two basic units are Weight value “1” for output value of 1
A subtracter 1 configured by a basic unit 1 that takes as input the product multiplied by "and -1" and does not include a threshold processing section 4
a (according to this configuration, it operates to calculate the difference value between the output values of these two basic units 1), and the weight value ω assigned to the human power of these two basic units 1.
1. By setting ω and the threshold value θ1θ appropriately, the assignment process of the membership function having the function shape “Temperature is normal” as shown in FIG. 7 can be realized.

In addition, the process of assigning the output function of the truth value of the consequent membership function involves dividing the consequent membership function into small sections and determining the truth value of each section, as shown in Figure 1O(a). The value y is specified, and then, as shown in FIG. 10(b), the truth value output device 1b constituted by the basic unit 1 without the threshold processing section 4 is set to the number of truth values (n pieces). The weight value "1K" is assigned to the input of this truth value output device 1b (which will specify the reduction rate of the output value of the truth value of the consequent membership function, as described later). Here, as shown in Fig. 10(b), the truth values of the consequent membership functions related to the same type of control operation amount, such as the opening degree of valve A, are The truth value output device 1b is configured to output the sum of the truth values of the reduced membership functions for the assigned control operation amount. It turns out.

In the embodiment shown in FIG. 1O, the function to output the function pattern of the consequent membership function is assigned, but in the present invention, the corresponding truth value of the consequent membership function is calculated. It is also possible to adopt a configuration in which it is used as a result value, and if such a configuration is adopted, Figures 8 and 9, which are adopted to realize the function of calculating the truth value of the antecedent membership function, can be used. The configuration will be used.

When the assignment process of the truth value calculation function of the antecedent part membership function and the truth value output function of the consequent part membership function is completed, the user next
The basic unit I-1 of the second intermediate layer is used to assign the antecedent calculation function, and the third basic unit 1 of the intermediate layer is used to perform the consequent calculation function. Executes the process of assigning functions.

This antecedent part calculation function assignment process can realize various calculation functions by using the basic unit 1 in various ways, and the most common method is to select and output the minimum value of the input values. However, here, as shown in FIG. 11(a), the basic unit which takes as input the product multiplied by the weight value "l/manpower score" and does not include the dark value processing section 4 By providing the average value calculator 1c constituted by 1, the arithmetic function of taking the average value of input values is executed by assigning it as an antecedent part arithmetic function. Similarly, this process of assigning consequent part calculation functions allows various calculation functions to be realized by using various basic units (1).
Also, although the most commonly used method is to select and output the maximum value of the input values, here, as shown in Figure 11 (b), By providing the addition value summer output unit 1d configured by the basic unit 1 which takes the input value as input and does not include the threshold value processing section 4, the calculation function that takes the addition value of the input value can be assigned as the consequent part calculation function. Execute.

In this way, instead of using one basic unit 1, it is also possible to realize the antecedent part calculation function and the consequent part calculation function by using a part of the hierarchical network in the hierarchical network part 11. It is. That is, a part of the hierarchical network part in the hierarchical network section 11 is secured as a unit for realizing the antecedent part calculation function and the consequent part calculation function, and the predetermined value is It is also possible to adopt a method of realizing the subject part calculation function and the consequent part calculation function by setting fixed values that have been learned. If this method is adopted, complex antecedent part calculation functions and consequent part calculation functions can be freely assigned.

The calculated value output by this consequent part calculation function is basically:
It gives the reduction rate of the truth value of the corresponding consequent membership function, for example, as shown in FIG.
If the membership function "Set the opening degree of pulp A small" is expressed as ■, and the membership function "Set the opening degree of r valve A large" is expressed as ■, then the members of ■ According to the output value of the consequent calculation function associated with the ship function, the application value of the membership function "Set the opening degree of valve A small" is determined, and ■
According to the output value of the consequent calculation function associated with the membership function of , the application value of the membership function of "set the opening degree of valve A to be large" is determined.

Most commonly, the opening degree Y of valve A, which is a fuzzy inference value, is calculated by finding the center of gravity of the figure of the function sum of the reduced membership functions according to the following formula. There is.

When the assignment of the antecedent calculation function and the consequent calculation function is completed, the user next uses the basic unit l of the output layer of the hierarchical network unit 11 to calculate the Natsuyama function (truth value output) of the fuzzy inference value. A function for calculating control operation amount data using the output value of the device 1b) is executed.

When this fuzzy inference value pure white function assignment processing is determined by, for example, the above-mentioned center of gravity calculation, as shown in FIG.
Two centroid derived value calculators 1e configured by are prepared, and as a weight value for internal connection with the truth value output device 1b,
For one gravity center derived value calculator 1e, the weight increases at equal intervals from 0 to 1, for example, in increasing order starting from the minimum value of the control operation amount associated with the truth value output device 1b. At the same time, the values are assigned to the other center of gravity derived value calculator 1e in descending order starting from the maximum value of the control operation amount associated with the truth value output device 1b, for example, from 0 to -1. This is achieved by assigning weight values that decrease at equal intervals. Then, the final centroid value is determined by these two centroid derived value calculators 1e according to calculation means prepared separately from the hierarchical network unit 11.
is configured to be calculated using the output value of

That is, with this configuration, one center of gravity derived value calculator 1
For example, in the case where the truth value output device 1b has 6 units, e is y, -o・Ca+0.2・C Fuji0.4・C8+0.6
・Ca+0.8・C%+l'c& However, C outputs the truth value of the control operation amount, and in contrast, the other center of gravity derived value calculator 1e
is Y, -1-C, -0,8・C, -0,6・C5-0,
4.C4-0, 2.C3-〇.C6 are output, so using the output values Y, , Y of this centroid derived value calculator 1e, 1 Y← Y, -Y.

If you plan, Y+ Y閤C+ + Cz + Cs + Ca + CS +C
6, the center of gravity value explained in equation (13) above is obtained.

FIG. 14 shows that the hierarchical network unit 11 is
The functional arrangement for realizing the fuzzy control rules that will be photographed in this figure is illustrated. Here, in the figure, the O mark without a symbol indicates the basic unit) 1. Also, the X in the diagram
I(SA) represents the truth value of Xl determined by the membership function SA for the control state quantity XI, LH5-1 represents the calculation result value of the antecedent part calculation, and Y(
SA) indicates that it is associated with the membership function SA for the control operation amount Y.

FIG. 15 shows a functional block diagram of FIG. 14. i! 1, those that perform the same functions as those explained in FIG. 1 are indicated by corresponding numerical values, such as the processing unit 13a corresponding to the processing unit 13 in FIG. 1. However, in the configuration shown in Figure 1,
The processing unit 16 of the fourth processing layer manages individual membership functions related to the same control operation amount type, and
While the output unit 17 is configured to calculate the fuzzy inference value by calculating the function sum of these membership functions, in this FIG. 15, the processing unit 16a of the fourth processing layer is Since this shows a configuration in which the function sum of membership functions related to the same control operation amount type is calculated at the same time, in FIG. In contrast, in FIG. 15, one processing unit 16a is provided for a group of processing units 15a related to the same control operation amount type.

In this way, as shown in FIG. 15, the user uses the intermediate layer at the forefront of the hierarchical network section 11 (in some cases, the next layer is used as shown in FIG. 14, but this is not an integral In place of the basic unit 1-i of the intermediate layer (which can be said to be the first-stage intermediate layer), a processing unit 13a that calculates and outputs the truth value of the membership function for the generated control state quantity is assigned, In place of the basic unit l of the middle layer located at the second position, a processing unit 14a that calculates and outputs the average value of human power is allocated, and in place of the basic unit 1 of the middle layer located third, A processing unit 15a that calculates and outputs the sum of the values is assigned, and instead of the basic unit 1 of the intermediate layer located at the fourth position, the truth value of the membership function for the generated control operation amount is assigned to the previous stage. Processing unit 15a located
Processing unit 1 that reduces and outputs according to the output value from
6a, and in place of the basic unit l of the output layer, an output unit (17a) that calculates and outputs two values used for calculating the center of gravity of the control operation amount,
Mapping of fuzzy control rules to the hierarchical network unit 11 is realized.

In this fuzzy control rule mapping process, the processing unit 13a (encircled by a broken line in the figure) calculates and outputs the truth value of the membership function of the same type of control state quantity, such as temperature, for example. Basically, a configuration is adopted in which internal connections are made only with specific identical input crab units 12 (those that distribute their control state quantity values), but this membership function is, for example, In the case of relating to different control state quantities such as an index, a configuration in which a plurality of input units 12 are internally coupled is adopted. Establishment of such an internal connection relationship between the input unit 12 and the processing unit 13a is achieved by setting zero weight values that are not changed by learning for other internal connections. Since the processing unit 14a executes the antecedent calculation of the fuzzy control rule, it is assigned several divisions of the fuzzy control rules to be generated, and in order to copy all possible combinations of the fuzzy control rules, Processing unit 13a
A configuration will be adopted in which all of the components and the processing unit 14a are internally coupled.

Furthermore, since the processing unit 15a executes the consequent computation of the fuzzy control rule, it is assigned to the number of membership functions of the control operation amount and copies all possible combinations of the fuzzy control rule. Therefore, a configuration is adopted in which the processing unit 14a and the processing unit 15a are all internally coupled. As explained in FIG. 10, the processing unit 16a and the corresponding processing unit 15a are all internally coupled, and as explained in FIG.
A configuration is also adopted in which all of the output units 17a and the output units 17a provided in 1-to-1 correspondence are internally coupled.

When the fuzzy control rule mapping process is completed, the user uses the system shown in FIG. By instructing the learning signal presentation device 20 to present the registered control data group to the hierarchical network configuration data processing bag f 10 a, the weight values of the hierarchical network unit 11 can be changed. Instruct the execution of learning.

In learning this weight value, the corresponding weight value (threshold value) explained in FIGS. 8 and 9 is set as the initial value of the weight value of the internal connection between the input crab unit 12 and the processing unit 13a. . In addition, as the initial value of the weight value of the internal connection between the processing unit 13a and the processing unit 14a, as explained in FIG.
is set. Further, as the initial value of the mesh value of the internal connection between the processing unit 14a and the processing unit 15a, "l" is set as explained in FIG. 11(b). Further, as the initial value of the weight value of the internal connection between the processing unit 15a and the processing unit 16a, a value that realizes the consequent membership function is set as explained in FIG. 1O.

Then, the corresponding weight value explained in FIG. 13 is set as the initial value of the weight value of the internal connection between the processing unit 16a and the output unit 17a.

When receiving control state quantity data for learning from the learning signal presentation device 20, the human crab unit 1 of the hierarchical network unit 11
2 transmits this control state prefix data to the corresponding processing unit 13.
Distribute to a. Subsequently, upon receiving this control state quantity data, each processing unit 13a calculates the truth value of the antecedent membership function and outputs the calculated truth value to the next stage processing unit 14a. Next, when the product value of the truth value of the antecedent membership function and the corresponding weight value is obtained, the processing unit 14a calculates the average value of the input values and sends it to the next stage processing unit 15a. Output as the calculated value of the antecedent part calculation. Subsequently, when the multiplication value of the calculated value of the antecedent part calculation and the corresponding weight value is obtained, the processing unit 15a
calculates the added value of the input values and outputs it to the next stage processing unit 16a as the calculated value of the consequent part calculation. continue,
When the multiplication value of the calculated value of the consequent part operation and the corresponding weight value is obtained, the processing unit 16a sends the output unit 17a a function of the truth value of the consequent part membership function reduced according to the human power value. Output the sum. Next, when we take the function sum of this consequent membership function, we get the output unit)17
a calculates and outputs the above-mentioned two values Y+ and Yz necessary for calculating the control operation amount data, and upon receiving this output, the arithmetic means (not shown) outputs the control operation amount data as the final output. is calculated and output to the weight value changing unit 30a and the learning convergence determining unit 23.

When the control operation amount data is received from the output unit 17a in this way, the weight value changing unit 30a performs the above-mentioned (7).
α and j are calculated according to the formula (8), and the update amount of the weight value of the internal connection between the output unit 17a and the processing unit 16a is calculated according to the formula (8). Here, the out crab knit 17a
Since the weight value of the internal connection between and the processing unit 16a does not have much meaning for generating fuzzy control rules, the weight value is not updated according to the calculated update amount. In addition, at the start of this learning process, the above-mentioned two Yl are used as teacher signals from the control operation amount data for learning.

It is required to calculate Y, but this process requires the connection between a straight line with a specific slope passing through the control operation amount data value for learning and the left and right end points of the consequent membership function. This will be carried out by a method such as determining according to the intersection points.

Next, the weight value changing unit 30a uses this calculated α□ to calculate β0 according to the above equation (9), and obtains (lO).
The update amount of the weight value of the internal connection between the processing unit 16a and the processing unit 15a is calculated according to the formula. here,
Since the weight value of the internal connection between the processing unit 16a and the processing unit 15a does not have much meaning for generating fuzzy control rules, the weight value is updated according to the calculated update amount. It doesn't have to be.

Subsequently, the weight value changing unit 30a changes the calculated β2. Calculate Tab according to the above equation (11) using
12) Processing unit) 15a and processing unit 1 according to formula
The update amount of the weight value of the internal connection with 4a is calculated. Since the weight value of the internal connection between the processing unit 15a and the processing unit 14a is a weight value that has meaning for generating fuzzy control rules, the weight value changing unit 30a
calculates a weight value for the next update cycle according to the calculated update amount and stores it in the weight value storage section 18.

Subsequently, the weight value changing unit 30a performs the same process as follows.
The update amount of the weight value of the internal connection between the processing unit 14a and the processing unit 13a, which has meaning for the generation of fuzzy control rules, is calculated, and the weight value for the next update cycle is calculated and the weight value is stored. It is stored in the section 18. Since the weight value of the internal connection between the processing unit 13a and the input crab unit 12 does not have much meaning for generating fuzzy control rules, the weight value changing unit 30a
Here, the weight value update process in this update cycle can be terminated.

If the learning convergence determination unit 23 of the learning signal presentation device 20 does not determine learning convergence of the weight values, the learning signal presentation unit 22 will present the same learning control data group again. The value changing unit 30a repeats the same procedure as the above process, thereby changing the value to the processing unit 15a.
The process of updating the weight value of the internal connection between the processing unit 14a and the processing unit 14a and the weight value of the internal connection between the processing unit 14a and the processing unit 13a continues.

Through this learning process, the weight value of the internal connection between the processing unit 15a and the processing unit 14a that expresses the relationship between the control data group for learning, and the weight value of the internal connection between the processing unit 14a and the processing unit 13a are determined. This weight value represents the actual control content of the controlled object. That is, an internal connection between a control state quantity and a control operation quantity that is defined as having a deeper relationship according to the actual controlled object has a larger internal state value.

From now on, processing unit 13a, processing unit 14a, and processing unit 15 connected by a weight value having a large value.
By extracting the relationship with a, it is possible to obtain a fuzzy control rule that is a faithful control model of the controlled object, as shown in FIG. 16, for example. Here, in FIG. 16, the processing unit 16a is shown in the same format as the description in FIG.

In the embodiments described above, the hierarchical network unit 11 made up of hardware components as shown in FIGS. 5 and 6 was disclosed. Next, an embodiment using the hierarchical network section 11 realized by a hierarchical network simulator built on a computer will be described in detail.

FIG. 17 shows the system configuration of a device used to implement the present invention using a hierarchical network simulator. In FIG. 17, the same parts as those explained in FIGS. It is indicated by the symbol. 10b is a hierarchical network simulator built on a computer. FIG. 18 shows a detailed functional configuration of this hierarchical network simulator 10b.

As shown in FIG. 18, the hierarchical network simulator 10b includes a processing unit 1 as shown in FIG.
3 a, 14 a + 15 a + 16 a *
and a network structure management section 40 that manages the connection relationship between the output unit 17a, an arithmetic control section 41 that controls the entire arithmetic processing, and each processing unit 13a, 14a, 1
5a and 16a; a weight value management unit 43 that manages weight values assigned to each inner connection; and an antecedent membership function, such as table means, etc. an antecedent membership function management section 44 that manages the consequent membership function using table means, a consequent membership function management section 45 that manages the consequent membership function using table means, etc., and each processing unit F13a+14.
Arithmetic execution unit 4 that executes the arithmetic functions of a, 15a, and 16a
6, and an output data management unit 47 that manages the calculated values that are the output values of each processing unit calculated by the calculation execution unit 46.
The hierarchical network simulator 10b executes a centroid calculation on the calculated value of the processing unit l-16a.
The center of gravity calculating section 48 calculates the control operation amount data that is the output value of.

Then, this calculation execution unit 46 is implemented by the processing unit 13 a
+ 14 a + 15 a + 16 a To simulate the process of + 14 a + 15 a + 16 a, a multiplication value calculation unit 49 calculates a multiplication value between a weight value and an input value, and an antecedent part membership function according to the multiplication value of the multiplication value calculation unit 49. The antecedent truth value calculation unit 50 calculates the truth value of , and the antecedent calculation function specifies the minimum value among the multiplication values of the multiplication value calculation unit 49 in case the antecedent calculation function calculates the minimum value of the input values. a minimum value specifying unit 51 that specifies the maximum value of the multiplication values of the multiplication value calculation unit 49, which is provided when the consequent calculation function calculates the maximum value of the input value; and a consequent truth value output unit 53 that reduces and outputs the function value of the consequent membership function.

The hierarchical network simulator 10b performs functions such as calculating an antecedent membership function, an antecedent calculation function, a consequent calculation function, and a consequent membership function according to the weight values and dark values assigned to the basic unit 1. Since it does not realize the output function of control state quantity data or the function of calculating the center of gravity of control state quantity data, unlike when using basic unit 1, the initial value of the weight value is set to l'', etc. (a random value is set). It is also possible to use discontinuous functional processing such as selecting the minimum value or maximum value as the antecedent part and consequent part calculation function. Therefore, when using such functional processing, the back propagation method described above cannot be used in a strict sense.
The present invention views the back propagation method in a broad sense, and adopts a configuration in which the internal state value learning processing device 30 executes learning of weight values according to the back propagation method. Of course, the application of the present invention is not limited to the method of learning weight values, so it is also possible to implement it according to a more appropriate learning algorithm. Also,
In this hierarchical network simulator 10b, since the consequent membership function management unit 45 directly manages the consequent membership functions by table means or the like, the processing unit 16a is configured as shown in FIG. As shown in the configuration, the configuration is provided in correspondence with the processing unit 15a.

Next, the processing of the present invention when using the hierarchical network simulator 10b will be explained.

When the control state quantity data for learning is received from the learning signal presentation device 20 to the input data development section 42, the multiplication value calculation section 49
reads out the input values of each processing unit 13a from the input data expansion section 42 in units of processing units 13a, reads out the weight values associated with those input values from the weight value management section 43, and performs their corresponding multiplication. Execute processing. When this multiplication value is calculated, the antecedent truth value calculation unit 50 reads out the corresponding antecedent membership function from the antecedent membership function management unit 44, and
The truth value associated with this multiplication value is calculated according to the read antecedent membership function and stored in the output data management section 47.

When the processing for all processing units 13a is completed, the output data management unit 47 expands the obtained truth value of the antecedent membership function to the input data expansion unit 42.

Subsequently, the multiplication value calculation unit 49 converts the input value of each processing unit 14a from the input data expansion unit 42 into the processing unit 14a.
are read out in units of input values, the weight values associated with these input values are read out from the weight value management section 41, and the corresponding multiplication processing is executed. When this multiplication value is calculated, the minimum value specifying unit 51 calculates the calculated value of the antecedent part calculation by specifying the minimum value among the multiplication values, and stores it in the output data management unit 47. All processing units) 1
When the processing for 4a is completed, the output data management section 4
7 expands the calculated result value of the antecedent part calculation to the input data expansion unit 42.

Subsequently, the multiplication value calculation section 49 converts the input value of each processing unit 15a from the human data development section 42 into the processing unit 15a.
are read out in units of input values, the weight values associated with these input values are read out from the weight value management section 43, and the corresponding multiplication processing is executed. When this multiplication value is calculated, the maximum value specifying unit 52 calculates the calculated value of the consequent part calculation by specifying the maximum value among the multiplied values, and stores it in the output data management unit 47. All processing units 1
When the processing for 5a is completed, the output data management section 4
7 expands the calculated result value of the consequent part calculation into the input data expansion unit 42.

Subsequently, the multiplication value calculation section 49 reads out the input values of the processing unit 16a from the input data development section 42 in units of processing units 16a, and receives the weight values associated with those input values from the weight value management section 43. At the same time as reading, these corresponding multiplication operations are executed. When this multiplication value is calculated, the consequent truth value output unit 53 reads out the corresponding consequent membership function from the consequent membership function management unit 45, and also reads out the consequent membership function that has been read out. The function value of is reduced according to this multiplication value and stored in the output data management section 47. All processing units 1
When the processing for 6a is completed, the output data management section 4
7 notifies the center of gravity calculation unit 48 of the function value of the reduced consequent membership function in units of control operation amounts of the same type, and upon receiving this notification processing, the center of gravity calculation unit 48 performs the above-mentioned (( 13) Calculate the control operation amount data that becomes the output value of the hierarchical network simulator lOb by executing the center of gravity calculation of equation Output to.

In this way the hierarchical network simulator tab
When the control operation amount data is received from the internal state 11 value learning processing unit f30, the internal state 11-value learning processing unit f30 executes processing similar to the processing of the weight value changing unit 30a described above to generate fuzzy control rules. A processing unit that is a meaningful weight value) 1
The update amount of the weight value of the internal connection between 5a and the processing unit 14a and the weight value of the internal connection between the processing unit 14a and the processing unit 13a is calculated, and the next update is performed according to the calculated update amount. The weight value for the cycle is calculated and the weight value of the weight value management unit 43 of the hierarchical network simulator 10b is updated.

If the learning convergence determination unit 23 of the learning signal presentation device 20 does not determine the learning convergence of the weight values, the learning signal presentation unit 22 will present the same learning control data group again. , internal state am learning processing device F30
By repeating this processing procedure, the learned values of the weight value of the internal connection between the processing unit 15a and the processing unit 14a and the weight value of the internal connection between the processing unit 14a and the processing unit 13a are obtained. learn.

Through this learning process, the hierarchical network unit 1 composed of hardware components as shown in FIGS. 5 and 6
1, the processing unit 15a and the processing unit 1) that express the relationship between the control data group for learning.
4a and the processing unit 14a.
The weight value of the internal connection between the processing unit 13a and the processing unit 13a is determined, and from this it is possible to obtain a fuzzy control rule that is a faithful control model of the controlled object, as shown in FIG. 16, for example. .

Next, the results of a simulation conducted to simulate the fuzzy control rule generation method according to the present invention will be described.

As shown in FIG.
Two pieces of each of 5a and 5a were prepared.

FIG. 20(a) shows the antecedent membership function assigned to the processing unit 13a-1, and FIG. 20(b) shows the antecedent membership function assigned to the processing unit 13a-1.
The calculation function of the antecedent membership function assigned to the processing unit 13a-2, FIG. 20(c), and the calculation function of the consequent membership function associated with the processing unit 15a-1, d), processing unit) 15
The calculation function of the membership function associated with a-2 is illustrated.

Here, the processing units 14a-1 and 14a-2 include
An antecedent part calculation function that calculates and outputs the average value of input values is assigned, and a consequent part calculation function that calculates and outputs an added value of input values is assigned to the processing units l5a-1 and 15a-2. Assigned.

Learning of weight values was performed assuming that the control data shown in FIG. 21 was obtained from the controlled object. The numerical values shown in FIG. 19 are the weight values of the learning results. It is clear that the internal connections indicated by the thick lines in the figure have large values. From now on, if X is B IG then Y is SM
ALLif X is SMAL L then Y
This means that the fuzzy control rule is B I G can be extracted.

Although the illustrated embodiment has been described, the present invention is not limited thereto. For example, the applicant previously
Japanese Patent Application No. 1983-227825 (filed on September 12, 1988, “Learning Processing Method for Network Configuration Data Processing Device”) aims to improve the back-propagating method to learn the weight values in a shorter time. Although the present invention has disclosed an invention that makes it possible to realize this, this learning method can be used in the present invention. Furthermore, in the embodiment, the explanation has been made by generating fuzzy control rules related to control state quantities as antecedent membership functions and control operation quantities as consequent membership functions, but the present invention is not limited to this. The present invention can also be directly applied to the generation of fuzzy control rules that relate to the control operation amount as the antecedent membership function and to control state quantities as the consequent membership function.

The antecedent part calculation function and the consequent part calculation function are not limited to those disclosed in the embodiments, and may take any calculation contents. Furthermore, any method may be used to implement the arithmetic processing of the processing unit and output unit, and although the fuzzy inference value obtained by centroid calculation has been disclosed, the present invention is not limited to this. Instead, a different fuzzy inference value calculation method may be used.

[Effects of the Invention] As explained above, according to the present invention, fuzzy control rules can be mechanically and automatically generated. Therefore, generation of fuzzy control rules, which was previously done manually, can now be performed objectively and with an extremely small amount of work. Fuzzy control rules can then be easily generated even when control knowledge regarding the controlled object is limited. Furthermore, even if a fuzzy controller is constructed according to the generated fuzzy control rules, it is possible to tune the fuzzy control rules to suit the control target using actual control data during operation. 0 or more, the present invention makes it possible to make the fuzzy controller more practical.

[Brief explanation of drawings]

Figure 1 is a diagram of the principle configuration of the present invention, Figure 2 is an explanatory diagram of the system used to realize the present invention, Figure 3 is a diagram of the basic configuration of the basic unit, and Figure 4 is the basic configuration of the hierarchical network section. Figure 5 is an example of the basic unit, Figure 6 is an example of the hierarchical network part, Figure 7 is an explanatory diagram of the membership function to be generated, and Figures 8 and 9 are the antecedent part membership. An explanatory diagram of the assignment process of the truth value calculation function of a function. Figure 1O is an explanatory diagram of the assignment process of the truth value output function of the consequent membership function. Figure 11 is an antecedent part calculation function and a consequent part calculation function. Fig. 12 is an explanatory diagram of fuzzy inference value calculation function assignment processing; Fig. 13 is an explanatory diagram of fuzzy inference value calculation function assignment processing; Figs. 14 and 15 are mapped fuzzy control diagrams. FIG. 16 is an explanatory diagram of extracted fuzzy control rules; FIG. 17 is an explanatory diagram of the system used to implement the present invention; FIG. 18 is an explanatory diagram of the hierarchical network simulator; FIG. 20 is an explanatory diagram of the hierarchical network part used in the simulation, FIG. 20 is an explanatory diagram of the membership function used in the simulation, and FIG. 21 is an explanatory diagram of learning control data used in the simulation. In the figure, ■ is the basic unit, 1 is the human crab unit, 2 is the multiplication processing section, 3 is the accumulation processing section, 4 is the threshold processing section, 10 is the hierarchical network configuration data processing means, and 10a is the hierarchical network configuration data processing device , 10b is a hierarchical network simulator, 11 is a hierarchical network unit, 12 is an input unit, 13, 14, 15 and 16 are processing units, 17 is an output unit, 18 is a weight value storage unit, 20 is a learning signal presentation device, 21 2 is a learning signal storage unit, 22 is a learning signal presentation unit, 23 is a learning convergence determination unit, and 30 is an internal state value learning processing device.

Claims (4)

[Claims] (1) A fuzzy control rule generation method for generating a fuzzy control rule that describes a control rule for a controlled object, which includes antecedent membership functions and consequent members that may constitute the fuzzy control rule. a first processing step (P1) that generates a ship function; and first, second, and third processing layers each including a plurality of processing units.
and a second processing layer, and between the second and third processing layers, hierarchical network configuration data comprising a hierarchical network unit (11) forming an internal connection in which internal state values are mutually set. A processing means (10) is prepared, a calculation function of the generated antecedent membership function is assigned to the processing unit (13) of the first processing layer, and the processing unit (14) of the second processing layer Allocate the antecedent calculation function of the fuzzy control rule for the number of fuzzy control rules to be generated,
A second processing step (P2) in which the consequent calculation functions of the fuzzy control rule are assigned as many as the generated consequent membership functions to the processing units (15) of the third processing layer.
Then, the control presentation data for learning is sent to the hierarchical network section (1
1), the control amount data is calculated from the output value from the third processing layer outputted according to the presentation and the generated consequent membership function, and the control amount data is calculated. A third processing step (P3
) and a fourth processing step (P4) in which fuzzy control rules are identified by analyzing the learned internal state values.
A method for generating a fuzzy control rule, comprising: (2) In the fuzzy control rule generation method according to claim (1), the second processing layer is configured to be further internally connected and configured hierarchically in multiple stages, and in the second processing step (P2), A method for generating a fuzzy control rule, comprising: assigning an antecedent calculation function of the fuzzy control rule to the second processing layer configured in multiple stages. (3) In the fuzzy control rule generation method according to claim (1), the third processing layer is configured to be further internally connected and configured hierarchically in multiple stages, and in the second processing step (P2), A method for generating a fuzzy control rule, comprising: assigning a consequent calculation function of the fuzzy control rule to the third processing layer configured in multiple stages. (4) In the fuzzy control rule generation method according to claim (1), (2) or (3), in the third processing step (P3), control teacher data for learning and control presentation data for learning are generated. Using the network output data of the hierarchical network unit (11) output in response to the presentation of A fuzzy control rule generation method characterized by performing processing such that learning of internal state values is performed as the process progresses.